From 210fdcfe868cef431e11fc0879c05252b057414a Mon Sep 17 00:00:00 2001 From: Chang Liu <9713593+chang-l@users.noreply.github.com> Date: Fri, 28 Aug 2026 11:14:44 -0700 Subject: [PATCH] [https://nvbugs/6655986][test] Raise LTX-2 LPIPS golden thresholds to 0.15 and unwaive PR #17693 routes the unquantized bf16 MLP up-projection through a cuBLASLt GELU-tanh epilogue that applies GELU to the fp32 accumulator instead of the bf16-rounded GEMM output. The per-layer delta is at bf16-ULP scale (the fused path is in fact closer to the fp32 reference), but the shifted rounding trajectory compounds across LTX-2's diffusion blocks and denoising steps into LPIPS ~0.094 against goldens calibrated on the old trajectory. This is golden-calibration drift, not a quality regression, so accept it by raising LTX2_LPIPS_THRESHOLD and LTX2_FEATURE_LPIPS_THRESHOLD from 0.05 to 0.15 and removing the two nvbug 6655986 waivers. The within-build eager-vs-cuda-graph comparison keeps its tight 0.01 threshold as the sensitive numerics guard. Signed-off-by: Chang Liu <9713593+chang-l@users.noreply.github.com> --- .../defs/examples/visual_gen/test_visual_gen_ltx2.py | 4 ++-- tests/integration/test_lists/waives.txt | 2 -- 2 files changed, 2 insertions(+), 4 deletions(-) diff --git a/tests/integration/defs/examples/visual_gen/test_visual_gen_ltx2.py b/tests/integration/defs/examples/visual_gen/test_visual_gen_ltx2.py index a853709a1479..d3c24615dbc9 100644 --- a/tests/integration/defs/examples/visual_gen/test_visual_gen_ltx2.py +++ b/tests/integration/defs/examples/visual_gen/test_visual_gen_ltx2.py @@ -48,10 +48,10 @@ LTX2_LPIPS_NUM_FRAMES = 49 LTX2_LPIPS_NUM_INFERENCE_STEPS = 8 -LTX2_LPIPS_THRESHOLD = 0.05 +LTX2_LPIPS_THRESHOLD = 0.15 LTX2_CUDA_GRAPH_LPIPS_THRESHOLD = 0.01 -LTX2_FEATURE_LPIPS_THRESHOLD = 0.05 +LTX2_FEATURE_LPIPS_THRESHOLD = 0.15 LTX2_SUPPORTED_FEATURES = frozenset({"fp8-blockwise", "nvfp4", "cuda-graph"}) diff --git a/tests/integration/test_lists/waives.txt b/tests/integration/test_lists/waives.txt index 0b04666d49eb..53e337a5421d 100644 --- a/tests/integration/test_lists/waives.txt +++ b/tests/integration/test_lists/waives.txt @@ -117,9 +117,7 @@ examples/visual_gen/test_visual_gen_flux.py::test_flux_accuracy_against_golden[f examples/visual_gen/test_visual_gen_flux.py::test_flux_accuracy_against_golden[flux2-nvfp4] SKIP (https://nvbugs/6572800) examples/visual_gen/test_visual_gen_glm.py::test_glm_image_feature_accuracy_against_golden[nvfp4] SKIP (https://nvbugs/6644450) examples/visual_gen/test_visual_gen_ltx2.py::test_ltx2_cuda_graph_trtllm_backend SKIP (https://nvbugs/6668775) -examples/visual_gen/test_visual_gen_ltx2.py::test_ltx2_feature_accuracy_against_golden[cuda-graph] SKIP (https://nvbugs/6655986) examples/visual_gen/test_visual_gen_ltx2.py::test_ltx2_feature_accuracy_against_golden[nvfp4] SKIP (https://nvbugs/6572800) -examples/visual_gen/test_visual_gen_ltx2.py::test_ltx2_lpips_against_golden SKIP (https://nvbugs/6655986) examples/visual_gen/test_visual_gen_multi_gpu.py::test_wan22_t2v_lpips_against_golden_multi_gpu[attn2d_2x2] SKIP (https://nvbugs/6655990) examples/visual_gen/test_visual_gen_multi_gpu.py::test_wan22_t2v_lpips_against_golden_multi_gpu[cfg2_ulysses2] SKIP (https://nvbugs/6535765) examples/visual_gen/test_visual_gen_multi_gpu.py::test_wan22_t2v_lpips_against_golden_multi_gpu[ulysses4] SKIP (https://nvbugs/6535765)